Patchline 🦥 — AI-Powered GitHub Issue Triage with Local Gemma 3

Shravan Sharma — Tue, 20 Oct 2026 14:58:45 +0000

Hacktoberfest 2026 — Week 3: Slow and Steady

What I Built

GitHub issues are essential to open-source development, but reviewing them can become repetitive. Maintainers often need to read lengthy reports, assess urgency, identify related issues, and decide what to investigate next.

I built Patchline, an open-source, AI-powered GitHub issue triage workspace that helps automate this initial review.

Instead of manually reading every issue from scratch, developers can paste a public GitHub repository URL and get an organized, AI-assisted overview.

Key Features

AI-powered issue summaries: Turn lengthy GitHub reports into concise, actionable summaries.
Suggested priorities: Classify issues by urgency with an explanation of the reasoning.
Automatic categorization: Organize issues into meaningful categories.
Recommended next actions: Help maintainers decide what to investigate next.
Similar-issue detection: Surface potentially related reports using title similarity.
Search and filters: Quickly navigate analyzed issues by keyword, category, and priority.
Issue inspector: View AI-generated insights alongside the original GitHub report.
Command palette: Navigate the workspace with keyboard shortcuts.

The goal is simple: spend less time sorting issues and more time solving them.

Demo

Here's Patchline in action, analyzing real GitHub issues using a locally running Gemma model.

Source code: GitHub — Patchline

The demo shows real GitHub issue analysis, AI-generated insights, and the workspace's search and navigation features. The illustrative sample preview is used to demonstrate similar-issue detection and is separate from live model inference.

How It Works

Patchline combines a React-based developer workspace with a Python backend and locally running AI.

Tech Stack

Frontend: React, TypeScript, Tailwind CSS, Vite
Backend: Python, FastAPI
AI model: Google Gemma 3 4B
Local inference: Ollama
Repository data: GitHub API

Analysis Workflow

The user enters a public GitHub repository URL.
The FastAPI backend fetches open issues from GitHub and excludes pull requests.
Relevant issue content is sent to Gemma 3 through the local Ollama API.
Gemma generates summaries, classifications, priority rationales, and suggested next actions.
Patchline displays these insights in an interactive dashboard.
A separate title-similarity check identifies potentially related issue reports.

Patchline is a read-only analysis tool. It does not automatically edit GitHub issues, change labels, or perform actions on behalf of maintainers.

The Slow and Steady Payoff 🦥

One of the most repetitive tasks in open-source maintenance is the first pass through incoming issues.

Before deciding what to fix, maintainers need to understand each report, assess its importance, and determine what to do next.

Patchline automates much of that initial review.

For example, when I analyzed an issue from the FastAPI repository, Patchline generated a concise summary, categorized the report as documentation, suggested a low priority, and recommended reviewing the roadmap items with maintainers.

Rather than immediately reading the entire report, I could first review the AI-generated overview and then inspect the original issue for additional context.

Patchline also includes a similar-issues view to help identify reports that might deserve investigation together.

The sample preview demonstrates this capability with illustrative data, while real repository analysis uses actual GitHub issues.

The result is a more focused triage workflow that keeps maintainers in control.

Why Open Innovation Matters

I chose Gemma 3 4B running locally through Ollama because I wanted to build useful AI automation without requiring a paid inference API.

Using an open-weight model offers several advantages.

1. Accessible Local AI

After downloading the model, developers can run issue analysis on their own machines without paying for each inference request.

2. Developer Control

The source code and analysis pipeline are available for inspection and modification. Developers can experiment with prompts, classifications, and workflow improvements.

3. Local Inference

AI processing happens through Ollama on the developer's machine rather than through a paid hosted model API.

GitHub access still requires an internet connection to retrieve live repository issues.

4. Reproducibility

Other developers can clone Patchline, install its dependencies, download Gemma 3, and run the same workflow locally.

The project is open source under the MIT License.

For me, open innovation means making AI-powered developer tools easier to inspect, modify, and run independently.

Challenges and Lessons Learned

Working with Real GitHub Data

One challenge was understanding how GitHub's API represents issues.

During testing, the API returned many open items from the FastAPI repository, but most were pull requests. After filtering those out, only one actual issue remained for analysis.

This reinforced the importance of testing with real repository data rather than relying entirely on mock responses.

Handling Long Issue Reports

Some GitHub issues contain extensive descriptions, logs, and technical details.

I needed to keep the input sent to Gemma bounded while still making the original issue report accessible in the interface.

The issue inspector now presents the AI-generated summary alongside the original report for reference.

Keeping AI Recommendations Advisory

AI-generated priorities and suggested actions are useful starting points, but they are not guaranteed to be correct.

Patchline therefore keeps the original issue available and clearly communicates that maintainer review is required.

The goal is to support human decisions, not replace them.

Testing and Validation

I validated Patchline through automated checks and real integration testing.

Development validation included:

26 passing backend tests
13 frontend DOM interaction checks
Successful frontend production build
Real GitHub API integration
Real Gemma 3 inference through local Ollama

I also manually reviewed the dashboard, issue inspector, search, filtering, and navigation flows.

What's Next?

There are several improvements I'd like to explore:

Semantic duplicate detection using embeddings.
Configurable issue-priority rules.
More flexible batch analysis.
Support for additional local AI models.
Better tools for maintainers working with large repositories.

The long-term goal is to make issue triage less repetitive while preserving the transparency and human oversight that open-source projects need.

Conclusion

Hacktoberfest's Slow and Steady theme inspired me to automate a task that developers and maintainers repeatedly perform.

Patchline brings together GitHub issue retrieval, local Gemma inference, and a focused developer workspace to make initial issue triage more efficient.

Let AI handle the repetitive first pass. Let developers focus on building. 🦥

GitHub Repository: https://github.com/shhhravann/patchline

Demo Video: https://youtube.com/shorts/9JZjW5uMTZA

Built for the Hacktoberfest 2026 Open-Source AI Challenge — Week 3: Slow and Steady.

Hello DEV! 👋 Starting my Hacktoberfest journey 🚀

Shravan Sharma — Sat, 03 Oct 2026 17:23:14 +0000

Hey everyone! 👋

I'm Shravan, a CSE(Data Science) student who loves building things and figuring out how they work.
I've been exploring web development, DSA, AI, and lately I've been wanting to get more involved with open source.
So, Hacktoberfest 2026 feels like the perfect excuse to finally jump in. 🎃
I'm hoping to learn from the community, make some meaningful contributions, build a few things, and hopefully meet some awesome developers along the way.
First open-source journey starts today. Let's see where it goes. 🚀
_Happy Hacktoberfest everyone! _🎃

DEV Community: Shravan Sharma